[libc++] Optimize __hash_table::find for integral types

[philnik777]

------------------------------------------------------------------------------------------------
Benchmark                                                                    old             new
------------------------------------------------------------------------------------------------
std::unordered_map<int, int>::find(key) (existent)/0                    0.007 ns        0.007 ns
std::unordered_map<int, int>::find(key) (existent)/32                    1.36 ns         1.01 ns
std::unordered_map<int, int>::find(key) (existent)/1024                  1.31 ns         1.15 ns
std::unordered_map<int, int>::find(key) (existent)/8192                  1.50 ns         1.16 ns
std::unordered_map<int, int>::find(key) (non-existent)/0                0.321 ns        0.318 ns
std::unordered_map<int, int>::find(key) (non-existent)/32                1.34 ns         1.36 ns
std::unordered_map<int, int>::find(key) (non-existent)/1024              1.08 ns         1.11 ns
std::unordered_map<int, int>::find(key) (non-existent)/8192              1.15 ns         1.21 ns
std::unordered_set<int>::find(key) (existent)/0                         0.007 ns        0.007 ns
std::unordered_set<int>::find(key) (existent)/32                         1.42 ns         1.07 ns
std::unordered_set<int>::find(key) (existent)/1024                       1.32 ns        0.962 ns
std::unordered_set<int>::find(key) (existent)/8192                       1.51 ns         1.17 ns
std::unordered_set<int>::find(key) (non-existent)/0                     0.325 ns        0.328 ns
std::unordered_set<int>::find(key) (non-existent)/32                     1.43 ns         1.31 ns
std::unordered_set<int>::find(key) (non-existent)/1024                   1.37 ns         1.26 ns
std::unordered_set<int>::find(key) (non-existent)/8192                   1.33 ns         1.21 ns
std::unordered_multimap<int, int>::find(key) (existent)/0               0.007 ns        0.007 ns
std::unordered_multimap<int, int>::find(key) (existent)/32               1.36 ns         1.05 ns
std::unordered_multimap<int, int>::find(key) (existent)/1024             1.45 ns        0.908 ns
std::unordered_multimap<int, int>::find(key) (existent)/8192             1.37 ns        0.922 ns
std::unordered_multimap<int, int>::find(key) (non-existent)/0           0.332 ns        0.344 ns
std::unordered_multimap<int, int>::find(key) (non-existent)/32           1.30 ns         1.28 ns
std::unordered_multimap<int, int>::find(key) (non-existent)/1024         1.18 ns         1.39 ns
std::unordered_multimap<int, int>::find(key) (non-existent)/8192         1.05 ns         1.33 ns
std::unordered_multiset<int>::find(key) (existent)/0                    0.007 ns        0.007 ns
std::unordered_multiset<int>::find(key) (existent)/32                    1.29 ns        0.981 ns
std::unordered_multiset<int>::find(key) (existent)/1024                  1.28 ns         1.05 ns
std::unordered_multiset<int>::find(key) (existent)/8192                  1.33 ns         1.22 ns
std::unordered_multiset<int>::find(key) (non-existent)/0                0.332 ns        0.332 ns
std::unordered_multiset<int>::find(key) (non-existent)/32                1.47 ns         1.29 ns
std::unordered_multiset<int>::find(key) (non-existent)/1024              1.09 ns         1.11 ns
std::unordered_multiset<int>::find(key) (non-existent)/8192              1.34 ns         1.21 ns

[ldionne]

I'm not super convinced. I find the optimization a bit ad-hoc and the results are not super significant. I think I'd rather explore the other optimization you mentioned where you don't store the hash at all in the node if the hash function desugars to identity.